2 Mar 2026
The evolution of artificial intelligence is entering a new phase. After the rapid adoption of generative AI tools focused largely on content creation, the next leg of the cycle is centred on Agentic AI systems that can autonomously plan, execute, and optimise multi‑step tasks. This transition has raised concerns around productivity disruption, IT spending, and employment, particularly in the software and services industry. However, a closer look at historical data and current trends suggests a more measured and structurally positive outlook.
From Generative to Agentic AI: A Structural Shift
Agentic AI represents a shift from AI systems that merely respond to prompts to systems that can act on behalf of users. This evolution may move AI up the value chain from content generation to operational optimisation and productivity enhancement, and eventually towards physical AI.
Importantly, such paradigm shifts in computing tend to be structural but slow. For context, despite nearly 25 years of adoption, e‑commerce penetration in the US remains at only ~16% of total retail sales*. Similarly, technologies once declared obsolete—such as mainframes—continue to be critical. Today, ^mainframes handle ~70% of global transactions, power ^86% of the world’s top 50 banks, and support ^80% of leading payment firms. These examples underline that new technologies typically layer on top of existing systems rather than replace them abruptly.
*Source: U.S. Census Bureau via Federal Reserve Economic Data (FRED), As per latest available data
^Source: https://www.ibm.com/downloads/documents/us-en/10c31775c85402a2
Productivity Gains Without Immediate Demand Destruction
A key promise of Agentic AI is its ability to compress multi‑step workflows into fewer actions. By eliminating intermediate steps and automating decision‑making, Agentic AI can materially improve productivity. However, productivity improvements do not automatically translate into lower overall spending.
Historical technology cycles demonstrate the relevance of Jevons’ Paradox where efficiency gains lower unit costs but expand total demand. This pattern has been observed across steam engines, electricity, enterprise software, the internet, and cloud computing. As the unit cost of intelligence declines, the number of economically viable applications tends to increase, often leading to higher aggregate technology spending over time.
AI Agents: A Large and Expanding Market
The scale of adoption anticipated for AI agents is significant. Estimates suggest that the AI agents market could expand from ~$5 billion in 2024 to nearly $300 billion by 2035. Over the same period, the number of AI agents in use could rise from ~255 million to over 15 billion, assuming an average pricing of ~$20 per month. This represents a multi‑year growth runway driven by enterprise adoption rather than consumer experimentation.
Such growth indicates that AI adoption is unlikely to be confined to a narrow set of use cases. Instead, AI agents are expected to proliferate across applications, workflows, and industries.
Source: MarketsandMarkets, Bernstein analysis and estimates dated 30th Apr 2025
Implications for the IT Services Industry
Contrary to fears of disruption, the data points towards a rising role for IT services companies in an AI‑driven world. AI systems are inherently probabilistic, with documented issues around hallucinations and limited reasoning at higher levels of complexity. Academic research highlights that even advanced reasoning models experience accuracy collapse beyond certain thresholds, reinforcing the need for governance, validation, and guardrails.
As a result, services firms may play a critical role in:
- AI governance and security frameworks
- Output validation and guardrails
- Workflow re‑engineering
- Integration of AI agents into legacy systems
This is reflected in global tech spending trends. IT services’ share of overall technology spend has steadily increased, while spending on devices and communication services has declined. By 2028, IT services are estimated to account for over *32% of global tech spending, compared to ~27% in 2019
* Source: Gartner estimates, Bernstein estimate & analysis, As on 14th July 2025.
New Revenue Pools for Services Firms
AI adoption is also creating incremental opportunity pools rather than displacing existing ones. Key areas include:
- Data centre upgrades, particularly the transition from CPU‑centric to GPU‑centric infrastructure
- Cloud and data migration
- Application re‑engineering, converting traditional software into AI‑enabled agents
- Building smaller, domain‑specific language models
- Process optimisation, RAG implementation, and workflow redesign
These areas are execution‑intensive and require deep domain expertise, favouring established IT services providers.
Revenue Models Are Already Evolving
Concerns around revenue deflation due to AI‑led productivity gains are also mitigated by the industry’s evolving revenue mix. Currently, *55–60% of IT industry revenues come from outcome‑based or fixed‑price contracts, rather than pure time‑and‑material billing. This structure allows firms to benefit from productivity improvements rather than be penalised by them.
Moreover, Indian IT firms are deeply embedded in product engineering for cutting‑edge global technologies, including enterprise AI platforms, cloud infrastructure, smart devices, and Indic language LLMs indicating a move beyond traditional labour arbitrage models.
*Source: Company Financials, Annual Report FY24-25, https://www.infosys.com/investors/reports-filings.html. As per latest available data.
Employment Impact: Data vs Perception
Job displacement fears have accompanied every major technology transition, and AI is no exception. However, recent data suggests that these concerns may be overstated.
- *Only ~16% of a software engineer’s time is spent on coding, with the remainder allocated to testing, security, deployment, user experience, and process management
- ^~80% of developers already use AI tools, yet utilisation rates at large IT firms have remained broadly stable
- Software development job postings have increased over the past five months
- Tech unemployment rates have stayed relatively stable over the last two years
*Source: IDC Report, Bernstein Analysis
^Source: Stack Overflow developer survey 2025, Bernstein Analysis
Further, survey data indicates that occupations with higher AI exposure are often seeing demand expansion rather than contraction, and divisions most affected by AI are reporting the least job displacement.
Sudheer Guntupalli, Vice President, Equity Research at Kotak AMC, adds: The neural networks powering LLMs work very similar to those in the human brain. As a technology, AI learns with more data / feedback (how stimulus affects the brain). In the last 3.5 years since launch, foundational models had gone through an exponential learning curve at general purpose applications like language, coding (language in a way), media creation etc. because billions of users are prompting the models, giving them generic stimulus on a daily basis. This is analogous to a child learning language and general purpose tasks at an exponential pace by the time he or she turns 5 years old.
But for the same child to master specific, complex and contextualised real world applications, it is going to take decades more of training, learning and hand holding. Enterprise scale up of AI should more or less take a similar path given enterprise problems, work flows and systems are more specific and customised to their business, rather than generic.
Disclaimers
The stocks/sectors mentioned do not constitute any recommendation and Kotak Mahindra Mutual Fund may or may not have any future position in these sectors/stocks. Use of the company brand names does not imply any affiliation with or endorsement by them or any of its holding companies, subsidiaries or affiliates and are used for illustrative purpose only. KMAMC is not guaranteeing/offering/communicating any indicative yield/returns on investments. The stocks/sectors mentioned do not constitute any kind of recommendation and are for information purpose only. These materials are not intended for distribution to or use by any person in any jurisdiction where such distribution would be contrary to local law or regulation. The distribution of this document in certain jurisdictions may be restricted or totally prohibited and accordingly, persons who come into possession of this document are required to inform themselves about, and to observe, any such restrictions. Kotak Mahindra Mutual Fund may or may not hold position in the mentioned stock(s)/sector(s).
Mutual Fund investments are subject to market risks, read all scheme related documents carefully.
